Mobile Robot Recognition Using Bayesian Penalization with Neural Approach

Abstract

We present in this paper a Bayesian classifier, based on neural probabilistic approach using radial basis function (RBF) and based on an improved version of orthogonal least square algorithm (OLS) for fast and incremental learning and automatic creation of hidden neurons. Applied to the famous case like inside a building, this classifier must assure a semantic localization, established on a realistic approach. The will wish to have a discrimination approach in the most possible case by using a generic and powerful representation of knowledge based on conditional and priori probabilities, error costs - case of decision throws etc., this classifier have been generated by neural network. Therefore in place to have a binary decision such as the hard decision like impasse, the mobile robot decides for example 90% of impasse situation.